TAJO-1279: Cleanup TajoAsyncDispatcher and interrupt stop events.

[jinossy]

No description provided.

[jihoonson]

I'm reviewing the patch.

[jihoonson]

When I run 'mvn clean install', tests are frequently hanged with the following error.

2015-01-07 21:58:07,530 INFO: org.apache.zookeeper.ClientCnxn (logStartConnect(975)) - Opening socket connection to server xxx.xxx.xxx.xxx/xxx.xxx.xxx.xxx:58986. Will not attempt to authenticate using SASL (unknown error)

It seems that the zookeeper client cannot find the proper server address.
Would you check this, please?

[jinossy]

@jihoonson
Thank you for the review!
Your log looks like security information. Could you show the error log to me ?
In my log, the connection was established successfully

2015-01-07 22:31:46,996 INFO: org.apache.zookeeper.ZooKeeper (<init>(438)) - Initiating client connection, connectString=kimjh.localhost:61503 sessionTimeout=90000 watcher=catalogtracker-on-hconnection-0x3dbc3890, quorum=kimjh.localhost:61503, baseZNode=/hbase
2015-01-07 22:31:46,998 INFO: org.apache.zookeeper.ClientCnxn (logStartConnect(975)) - Opening socket connection to server kimjh.localhost/192.168.101.41:61503. Will not attempt to authenticate using SASL (unknown error)
2015-01-07 22:31:46,998 INFO: org.apache.zookeeper.ClientCnxn (primeConnection(852)) - Socket connection established to kimjh.localhost/192.168.101.41:61503, initiating session
2015-01-07 22:31:46,998 INFO: org.apache.zookeeper.server.NIOServerCnxnFactory (run(197)) - Accepted socket connection from /192.168.101.41:65438

[hyunsik]

+1
The patch looks good to me. The unit tests are successfully passed. The patch seems to find unreleased resources.

@jihoonson,

Do you have more stuff that you need to review? If not, I want that this patch is committed to master as soon as possible due to recently frequent CI failures.

[jihoonson]

@hyunsik thanks for your review. I also think that we should resolve this issue as soon as possible.

All changes look good to me, too. The only thing I concern is, as commented above, the frequent hangs during unit tests. Whenever a zookeeper client fails to establish a connection, the test waits for 10 ~ 50 seconds. I think that it will significantly bother us.

I cannot find this error in the current master branch. Thus, I suspect that an existing, but hidden error is exposed due to the changed HBase-related configurations. Maybe we can fix it in another issue.

What do you think about it?

[jinossy]

@jihoonson I will remove the HBase-related codes

+    conf.setInt(HConstants.HBASE_CLIENT_IPC_POOL_SIZE, 5);
+    conf.setInt(HConstants.REGION_SERVER_HANDLER_COUNT, 5);
+    conf.setInt(HConstants.REGION_SERVER_META_HANDLER_COUNT, 2);
+    conf.setInt(HConstants.MASTER_HANDLER_COUNT, 5);
+    conf.setInt("hbase.hconnection.threads.max", 5);
+    conf.setInt("hbase.hconnection.threads.core", 5);

[jihoonson]

@jinossy thanks. We can commit the patch immediately.
Anyway, we must resolve the zookeeper-related problem in another issue.

[jinossy]

@jihoonson
I don’t think so. Hbase was merge by TAJO-1233. and This patch is not related miniZookeeperCluster binding issue.
Do you still have the problem in this patch?

[jihoonson]

@jinossy, I have tested the latest patch. It worked well.
Here is my +1.
Thanks!

[jinossy]

@jihoonson Thanks!
I'll commit it shortly.